Skip to main content
Pure Technical Services

Quick Reference: Best Practice Settings

Currently viewing public documentation. Please login to access the full scope of documentation.

KP_Ext_Announcement.png

This serves as a quick reference for VMware vSphere ESXi hosts that are integrated with a Pure Storage FlashArray.

For additional in-depth information surrounding best practices please refer to Pure Storage KB: Web Guide: FlashArray VMware Best Practices.

Best Practices for ALL versions of ESXi

ESXi Parameters Recommended Default Description

HardwareAcceleratedInit

1

1

Enables and controls use of Block Same (WRITESAME).

HardwareAcceleratedMove

1

1

Enables and controls use of XCOPY.

VMFS3.HardwareAcceleratedLocking

1

1

Enables and controls the use of Atomic Test & Set (ATS).

iSCSI Login Timeout

30

5

Ensures iSCSI sessions survive controller reboots.

TCP Delayed ACK (iSCSI) Disabled Enabled Improves performance when disabled in congested networks

Jumbo Frames (Optional)

 

9000 1500 If you have a workload that would benefit from Jumbo Frames, and a network that supports it, then this is the recommended configuration. Otherwise, utilize 1500 to reduce complexity in configuration.
HBA Queue Depth Limits Default Varies by vendor Default is recommended unless specifically requested by Pure Storage due to high-performance workloads.

VMware Tools

Install

Not Installed

VMware paravirtual driver, clock sync, increased disk timeouts, and graphic support are part of the tools hence it is a crucial step.

VM Virtual SCSI Adapter Paravirtual Varies by OS type Not required to be changed, but for high-performance requirements, PVSCSI is required to be used.
Network Time Protocol (NTP) Enabled Disabled Enabling NTP is recommended for more efficient troubleshooting.
Remote Syslog Server Enabled Disabled Configuring a remote syslog server is recommended to ensure logging required for troubleshooting is available.

It is also a best practice to set the "ESXi host personality" on the FlashArray for all ESXi host objects. This is described in detail in: FlashArray Configuration - Setting the FlashArray ESXi host Personality section. Please ensure this is applied whenever possible.

Best Practices specific to ESXi 5.x

ESXi Parameters Recommended Default Description
DSNRO / Number of outstanding IOs 32 32 Default is recommended unless specifically requested by Pure Storage due to high-performance workloads.
Path Selection Policy Round Robin MRU Path Selection Policy for FC and iSCSI.
IO Operations Limit (Path Switching) 1 1000 How many I/Os until ESXi switches to another path for a volume.
VMFS Version 5 5 Please upgrade if any VMFS-3 datastores are in use.
UNMAP Block Count 1% of free VMFS space or less 200 Please refer to the VAAI or Best Practices document for additional information.

Disk.SchedNumReqOutstanding (DSNRO) is the same as "Number of outstanding IOs". The difference is that it changed from a host level configuration (Disk.SchedNumReqOutstanding) to a per volume level (Number of outstanding IOs) in ESXi 5.5 and later.

You can read the VMware KB Article: Setting the Maximum Outstanding Disk Requests for Virtual Machines for additional information around this.

Best Practices specific to ESXi 6.x and ESXi 7.x+

ESXi Parameters Recommended Default Description
EnableBlockDelete 1 0 Provides end-to-end in guest support for space reclamation (UNMAP). Only applicable to VMFS-5.

Number of outstanding IOs 
(Per LUN QDepth)

32 32 Default is recommended unless specifically requested by Pure Storage due to high-performance workloads.
Path Selection Policy Round Robin MRU Path Selection Policy for FC and iSCSI.
Latency Based PSP (ESXi 7.0+)

samplingCycles - 16
latencyEvalTime - 180000 ms

samplingCycles - 16
latencyEvalTime - 180000 ms
How often a path is evaluated (every 3 minutes) and how many I/Os to sample (16) during the evaluation.

IO Operations Limit (ESXi 6.0 - 6.7)

1 1000 How many I/Os until ESXi switches to another path for a volume.
VMFS Version 6 5 Use VMFS-6 on vSphere 6.5 and later.

If vSphere 6.7U1 or later are in use then you can use the VMW_PSP_RR module set to "latency" rather than IO Operations set to 1. In vSphere 7.0 and later, you should use VMW_PSP_RR module set to "latency". Please refer here for more information on Enhanced Round Robin Load Balancing and here for more details on why Pure recommends this change.